Dataset and Python Script - Low-Cost Condition Assessment of Unpaved Rural Roads for Sustainable Maintenance: Visual Indices and Smartphone-Based IRI in the Ecuadorian Andes

Overview This dataset contains the complete field database and the analysis code supporting the study of low-cost condition assessment of unpaved rural roads on the Las Chinchas–Zambi corridor (15 km, Loja Province, Ecuador). Four visual-manual condition indices (URCI, MTC, PASER and ICNP) were surveyed alongside two roughness measurement systems — a Class III response-type profilometer (Roughometer III) and a free smartphone application — within a single seven-day dry-season window. All variables were harmonised to 150 stations of 100 m. The two files provided allow every table and figure of the associated article to be reproduced from a single execution. Files 1. Consolidado resultados Zenodo.xlsx — raw and processed field database, organised in three main sheets: IRI_ROUGHMETER — International Roughness Index (IRI) recorded by the Roughometer III over four complete runs of the corridor (two outbound, two return), with chainage and operating speed. IRI_APP — IRI recorded by the smartphone application: five series obtained from four traverses, using two devices (Google Pixel 6 and OnePlus 6); on the final traverse both phones recorded simultaneously. Important: in each column block, the valid processed series occupies the upper rows (approximately rows 1–154, at 100 m intervals). Below them lies a consolidation and inversion staging area whose columns are shifted by one position relative to the block above. Reading a column continuously from top to bottom mixes IRI values with operating speeds. The analysis script delimits the valid series automatically. RESULTS — visual index surveys at their native sampling scales, the interpolated grid, and the DEFINITIVE block containing the six harmonised variables used in the models. Note: in this block the column labelled DNV(ICPN) corresponds to the ICNP index. 2. unpaved_road_analysis_Zenodo.txt — the complete analysis script (Python). It is distributed with a .txt extension for platform compatibility and must be renamed to .py before execution. How to use Download both files into the same folder. Rename unpaved_road_analysis_Zenodo.txt to unpaved_road_analysis.py. Open the script and check that the input filename declared at the top matches the name of the Excel file as downloaded, and that the output folder path suits your system. These are the only two lines that need editing. Install the dependencies (see below). Run the script. It creates an output folder containing a results workbook and the figures of the article, in PNG format at 300 dpi. Requirements. The analysis was developed and executed in Python 3.13.9 (64-bit) under the Spyder IDE on Windows 11 (AMD64), using pandas 2.3.3, NumPy 2.3.5, SciPy 1.16.3 and Matplotlib 3.10.6, together with statsmodels (Lilliefors normality test) and scikit-learn (linearly weighted kappa). Other recent versions of these libraries are expected to work. Random resampling uses a fixed seed, so bootstrap confidence intervals are exactly reproducible. What the script computes Descriptive statistics of the harmonised variables; repeatability of the profilometer (intraclass correlation coefficients, coefficient of variation, Bland–Altman agreement, direction and speed effects); structure, spatial consistency and screening capacity of the smartphone series (Kendall's W, ICC on standardised scores, quartile agreement and weighted kappa, sensitivity to the spatial aggregation window); normality of all variables reported with both the uncorrected Kolmogorov–Smirnov p-value and the Lilliefors p-value; six regression models with leave-one-out cross-validation, prediction errors, bootstrap confidence intervals of R² and full residual diagnostics (Shapiro–Wilk, Breusch–Pagan and Durbin–Watson); robustness of the models recomputed at each index's native sampling scale without interpolation; and the complete Spearman correlation matrix. Two analyses reported in the article are not reproduced by this script: the Passing–Bablok non-parametric regression, computed with the Real Statistics add-in for Microsoft Excel, and the comparison of alternative functional forms, performed in Statgraphics. Scope and reuse The data describe a single Andean corridor surveyed in one dry season. The regression equations are therefore site-specific, and their transfer to other corridors or to wet-season conditions requires additional validation. The database is nonetheless suitable for methodological reuse: comparison of visual condition indices, studies of measurement repeatability on unpaved surfaces, and evaluation of smartphone-based roughness estimation. Citation: Manuscript under review